AI Agent Engineer Job Description Template

Why most agent job postings attract the wrong people

Read ten job postings for AI engineers and eight of them will be effectively interchangeable. They open with a paragraph about the company's mission, list a stack of technologies, ask for three to five years of experience with large language models, and mention that the candidate should be passionate about artificial intelligence. Nothing in that posting distinguishes an engineer who has recovered a looping agent at midnight from one who has completed a tutorial series and built a weekend project.

The consequence is entirely predictable. You receive two hundred applications, most of them from people whose relevant experience is a side project, and your screening cost balloons accordingly. Worse, the strongest candidates frequently do not apply at all, because a generic posting signals a team that has not yet worked out what it actually needs — which is precisely the engagement that experienced agent engineers have learned to avoid.

The fix is uncomfortable but reliably effective: describe the problem, including the parts that are slightly embarrassing. A posting that says 'our support agent works well until a tool times out, at which point it retries indefinitely and we cannot tell why from our logs' will receive fewer applications and dramatically better ones, because it is a problem a practitioner recognises immediately.

How to structure the posting

1. 01

   Open with the failure, not the mission

   The first paragraph should describe the specific problem this hire will own, in concrete terms. Practitioners scan for exactly this and stop reading when it is absent. Company mission statements belong further down the page, where they inform a candidate who is already interested rather than filtering out the ones who would have been.

2. 02

   List outcomes rather than technologies

   Write 'you will build the evaluation harness that gates every prompt change' instead of 'experience with evaluation frameworks required'. Outcomes tell a candidate what their actual week looks like and let them self-assess honestly, which removes an entire screening round for both sides of the process.

3. 03

   Separate must-have from nice-to-have honestly

   Most postings list twelve requirements and genuinely mean four. Be explicit about which is which. Production agent experience and strong general engineering are usually the real must-haves; specific frameworks, cloud providers and domain familiarity almost always belong in a clearly labelled nice-to-have section.

4. 04

   State the rate range

   Omitting the range costs you strong candidates who assume the worst, and it filters in candidates who are optimising for something other than compensation. It also removes several rounds of mutual time-wasting in the cases where expectations were never going to overlap at all.

5. 05

   Say what the first ninety days look like

   Experienced agent engineers want to know whether they will be permitted to add tracing and build an evaluation harness before being asked to improve output quality. Saying so explicitly is a genuinely strong differentiator, because the overwhelming majority of postings do not address it.

Requirements worth cutting entirely

• A specific degree — this field is far too young for formal credentials to mean very much
• Certifications in any AI platform, which correlate weakly with actual production capability
• A minimum number of years working with language models, which excludes strong recent converts
• Long lists of frameworks, most of which represent a few days of ramp-up for a strong engineer
• 'Passion for AI', which every applicant will claim and against which none can be assessed
• Publication or conference speaking history, unless the role genuinely involves external work
• Willingness to work in a fast-paced environment, which communicates nothing to anybody

What the strongest candidates read the posting for

Experienced agent engineers are, in practice, screening you as hard as you are screening them. The field is small, demand exceeds supply, and a bad six-month engagement costs them more than it costs you. They read a posting looking for evidence that the team has thought seriously about the problem, and they draw quite reliable inferences from what is absent.

Three absences are read as warnings. No stated problem suggests the agent is a mandate from above rather than a response to a need. No mention of evaluation suggests the team will judge the work on demos and vibes, which makes the engineer's contribution unmeasurable and their tenure precarious. No mention of what already exists suggests either a greenfield fantasy or an undisclosed legacy system, and experienced people have been surprised by both.

Addressing all three costs you perhaps a hundred words. It is the single highest-return edit available on a job posting for this role, and it is the one most teams skip because it requires admitting publicly that something in the current system does not work.

Part of the Hiring AI Agent Engineers cluster · Read the pillar page